BenchEval is a physician-led clinical AI evaluation company focused on testing high-stakes AI, beginning with medicine. Founded by emergency physician Dr Omer Atli and neurologist Dr Semsettin Atli, the company works with AI labs and healthcare AI teams to identify where models fail, especially when confident but unsafe responses could cause harm.

Its services include blinded, rubric-driven evaluations of model performance on clinical tasks; adversarial benchmarks built around difficult, atypical and time-pressured cases; physician-led red-teaming for problems such as missed escalation, unsafe reassurance and incorrect dosing; and specialty-matched post-training data, including demonstrations, graded comparisons and corrections. BenchEval’s published Standard defines grading dimensions, failure categories and severity levels before testing begins. Its process combines specialty-matched clinician calibration, blinded review by a second physician, and adjudication of disagreements.

Clients receive graded transcripts, clinical rationales, versioned task sets and reproducible findings ranked by clinical severity. BenchEval presents results as evidence about observed behaviour under tested conditions, not as certification, regulatory approval or a guarantee of safety.